HELM v2 Lite - LegalBench: leaderboard
The "LegalBench" subtask of HELM v2 Lite. Stanford HELM Lite v2 — lightweight HELM run comparing frontier models on core scenarios with exact-match-style metrics.
Metric: Exact Match (%). Source: nlp.stanford.edu. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 (0613) | 71.01 | #491 |
| 2 | Claude 2 | 61.17 | #761 |
| 3 | Llama 2 70B | 60.58 | #1201 |
| 4 | Claude 2.1 | 60.11 | #839 |
| 5 | GPT-4 Preview (1106) | 59.26 | #418 |
| 6 | GPT-3.5 Turbo (0613) | 58.24 | #1118 |
| 7 | text-davinci-003 | 56.06 | #1146 |
| 8 | Claude v1.3 | 51.53 | #1001 |
| 9 | text-davinci-002 | 51.44 | #1225 |
| 10 | Yi 34B (Base) | 50.66 | #972 |
| 11 | Llama 2 13B | 48.51 | #1436 |
| 12 | LLaMA-65B | 46.01 | #1303 |
| 13 | Yi 6B (Chat) | 44.46 | #1408 |
| 14 | Mistral-7B-v0.1 | 44.24 | #1326 |
| 15 | falcon-40B | 40.09 | #1442 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-v2-lite-legalbench · How It Works · Data refreshed daily, snapshot 2026-10-20.